The use of probabilistic evolutionary models is an integral part of comparative genomics. Applications include design of evolutionary genefinding software;screening of coding DNA sequences for sites under selection;identification of regulatory elements in genomes;improved detection of homologues;and prediction of deleterious SNPs. Such evolutionary models remain under-developed, particularly for the subclass of genes expressed solely as noncoding RNA (ncRNA). However, there is increasing biomedical interest in this class of genes: microRNAs are now suspected to regulate a wide range of targets, and are used by viruses to silence host transcription;small interfering RNAs are being explored as atherapeutic treatment;drugs exist that specifically target RNA structure (e.g. bacterial ribosomes, or retroviral binding sites such as HIV's Rev Response Element);RNA motifs known as "riboswitches", echoing genetically- engineered "aptamers" in their ability to discriminatively bind small-molecule ligands, have recently been discovered in prokaryotes;many catalytic ncRNAs ("ribozymes") are active in human and bacterial cells;and many of the above technologies (e.g. riboswitches/ aptamers/ ribozymes) are beginning to be exploited by synthetic biologists e.g. as novel reporter constructs, with great biomedical potential. Here, we propose to develop ncRNA evolutionary models for a focussed, biologically testable case study: the identification of ncRNA genes by comparative analysis of twelve fruitfly genomes. Computationally, we will use this example to drive forward our ongoing methods development in RNA analysis and evolutionary modeling, adapting our successful "xrate" and "stemloc" programs for use with evolutionarystochastic context-free grammars. Experimentally, we will test our predictions by wet-lab validation methods such as RT-PCR and sequencing, with the help of our collaboratorsand locally available resources such as the Berkeley Drosophila Genome Project's cDNA libraries. All our software will be freely available online.